drm/mgag200: Add a workaround for low-latency

We found a regression in v5.10 on real-time server, using the
rt-kernel and the mgag200 driver. It's some really specialized
workload, with <10us latency expectation on isolated core.
After the v5.10, the real time tasks missed their <10us latency
when something prints on the screen (fbcon or printk)

The regression has been bisected to 2 commits:
commit 0b34d58b6c32 ("drm/mgag200: Enable caching for SHMEM pages")
commit 4862ffaec523 ("drm/mgag200: Move vmap out of commit tail")

The first one changed the system memory framebuffer from Write-Combine
to the default caching.
Before the second commit, the mgag200 driver used to unmap the
framebuffer after each frame, which implicitly does a cache flush.
Both regressions are fixed by this commit, which restore WC mapping
for the framebuffer in system memory, and add a cache flush.
This is only needed on x86_64, for low-latency workload,
so the new kconfig DRM_MGAG200_IOBURST_WORKAROUND depends on
PREEMPT_RT and X86.

For more context, the whole thread can be found here [1]

Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>
Link: https://lore.kernel.org/dri-devel/20231019135655.313759-1-jfalempe@redhat.com/ # 1
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20240208095125.377908-1-jfalempe@redhat.com

3 files changed, 37 insertions, 0 deletions

diff --git a/drivers/gpu/drm/mgag200/Kconfig b/drivers/gpu/drm/mgag200/Kconfig
index b28c5e4828f4..5e4d48df4854 100644
--- a/drivers/gpu/drm/mgag200/Kconfig
+++ b/drivers/gpu/drm/mgag200/Kconfig
@@ -11,3 +11,15 @@ config DRM_MGAG200
 	 MGA G200 desktop chips and the server variants. It requires 0.3.0
 	 of the modesetting userspace driver, and a version of mga driver
 	 that will fail on KMS enabled devices.
+
+config DRM_MGAG200_IOBURST_WORKAROUND
+	bool "Disable buffer caching"
+	depends on DRM_MGAG200 && PREEMPT_RT && X86
+	help
+	  Enable a workaround to avoid I/O bursts within the mgag200 driver at
+	  the expense of overall display performance.
+	  It restores the <v5.10 behavior, by mapping the framebuffer in system
+	  RAM as Write-Combining, and flushing the cache after each write.
+	  This is only useful on x86_64 if you want to run processes with
+	  deterministic latency.
+	  If unsure, say N.
diff --git a/drivers/gpu/drm/mgag200/mgag200_drv.c b/drivers/gpu/drm/mgag200/mgag200_drv.c
index 54fce00e2136..573dbe256aa8 100644
--- a/drivers/gpu/drm/mgag200/mgag200_drv.c
+++ b/drivers/gpu/drm/mgag200/mgag200_drv.c
@@ -84,6 +84,20 @@ resource_size_t mgag200_probe_vram(void __iomem *mem, resource_size_t size)
 	return offset - 65536;
 }
 
+#if defined(CONFIG_DRM_MGAG200_IOBURST_WORKAROUND)
+static struct drm_gem_object *mgag200_create_object(struct drm_device *dev, size_t size)
+{
+	struct drm_gem_shmem_object *shmem;
+
+	shmem = kzalloc(sizeof(*shmem), GFP_KERNEL);
+	if (!shmem)
+		return NULL;
+
+	shmem->map_wc = true;
+	return &shmem->base;
+}
+#endif
+
 /*
  * DRM driver
  */
@@ -99,6 +113,9 @@ static const struct drm_driver mgag200_driver = {
 	.major = DRIVER_MAJOR,
 	.minor = DRIVER_MINOR,
 	.patchlevel = DRIVER_PATCHLEVEL,
+#if defined(CONFIG_DRM_MGAG200_IOBURST_WORKAROUND)
+	.gem_create_object = mgag200_create_object,
+#endif
 	DRM_GEM_SHMEM_DRIVER_OPS,
 };
 
diff --git a/drivers/gpu/drm/mgag200/mgag200_mode.c b/drivers/gpu/drm/mgag200/mgag200_mode.c
index 0eb769dd76ce..e17cb4c5f774 100644
--- a/drivers/gpu/drm/mgag200/mgag200_mode.c
+++ b/drivers/gpu/drm/mgag200/mgag200_mode.c
@@ -13,6 +13,7 @@
 
 #include <drm/drm_atomic.h>
 #include <drm/drm_atomic_helper.h>
+#include <drm/drm_cache.h>
 #include <drm/drm_damage_helper.h>
 #include <drm/drm_edid.h>
 #include <drm/drm_format_helper.h>
@@ -436,6 +437,13 @@ static void mgag200_handle_damage(struct mga_device *mdev, const struct iosys_ma
 
 	iosys_map_incr(&dst, drm_fb_clip_offset(fb->pitches[0], fb->format, clip));
 	drm_fb_memcpy(&dst, fb->pitches, vmap, fb, clip);
+
+	/* Flushing the cache greatly improves latency on x86_64 */
+#if defined(CONFIG_DRM_MGAG200_IOBURST_WORKAROUND)
+	if (!vmap->is_iomem)
+		drm_clflush_virt_range(vmap->vaddr + clip->y1 * fb->pitches[0],
+				       drm_rect_height(clip) * fb->pitches[0]);
+#endif
 }
 
 /*